Title: The Cloud Desktop Silicon Problem: The CPU Lottery Hiding Inside Your EUC Bill
Speaker: Kees Baggermann
I lead a team at Nutanix focused on developing proven methodologies for deploying and optimizing enterprise applications on the Nutanix platform. My responsibilities span the entire lifecycle—from gathering business and technical requirements, to designing architectures, to supporting successful customer implementations. Prior to joining Nutanix, I spent many years delivering complex infrastructure migrations and deployments across Microsoft and Citrix environments, including RDS, XenApp, XenDesktop, NetScaler, PVS, and frequently RES Workspace Manager. That hands-on experience continues to inform my approach to solving infrastructure challenges, ensuring solutions are both technically sound and operationally practical.
Abstract:
Everyone running virtual desktops knows their SKU. Almost no one knows which physical CPU is actually under it, and that gap is costing you money and performance right now.
This session breaks down why cloud desktop instances across Azure, AWS, and Google Cloud run on a silicon spread that can span nearly a decade, all sold at the same hourly price. We'll cover the economics behind it (hyperscaler depreciation schedules that make older hardware pure margin), the mechanics of it (why a single SKU like Standard_D4s_v5 can land you on anything from a 2019 Cascade Lake to a 2021 Ice Lake host, with no way to request a specific generation), and what it means for you in practice: a documented 26% performance spread for identical pricing, and re-roll odds that are worse than a coin flip.
You'll leave with a playbook: how to audit your fleet's actual silicon vintage with a single CIM query, how to use proximity placement groups and host pool segmentation to narrow your odds, when dedicated hosts are worth the premium for a guarantee, and why locking in reservations before mid-2026 pricing changes matters.
Takeaway: A SKU version number is a feature-set boundary, not a CPU-generation guarantee. Query the CPU, segment the pools, and stop letting the silicon choose for you.
